Application of sliding mode observers for actuator fault detection and isolation in linear systems

This paper investigates the actuator fault isolation problem for a class of linear systems. To isolate the actuator faults amongst all possibilities, we first relate each possibility to a faulty model, then for each possible faulty model, a sliding mode observer (SMO) is designed. It is proved that, for the faulty model corresponds to the faulty actuators, the SMO designed can ensure the related state estimation error and thus the output estimation error goes to zero. It is also shown that, for all other possible faulty models, none of SMOs can make the related output estimation errors be zero. Based on the results proved, we define the residuals as the square of the magnitudes of the output estimation errors resulting from all possible faulty models. If only residual goes to zero, then it corresponds to the faulty actuators, and actuator fault isolation is done. The use of SMOs has two advantages. One is that it can deal with any types of bounded actuator faults (constant and non-constant faults); the other is that it can provide a method to estimate the faults. The actuator fault isolation method is tested on a research civil aircraft model (RCAM), and simulation results show that it can isolate various types of actuator faults effectively